Weekly drop 2026-08-15: 30 added, 0 updated

[assafbar2]

401 → 431 agents. 30 added, 0 updated. Validation passes (431 agents, 20 categories, 0 violations).

Lookback window: 2026-08-10 (last drop) → 2026-08-15, 4 days, plus the mandatory 7-day gap sweep back to 2026-08-03. The short window made Mode A thin, so this run is weighted toward Mode B backlog — and the backlog turned out to be large.

Verification method: every star count below came from gh api repos/<owner>/<repo> at authoring time, never from memory or a scraped page. Every product URL was fetched with curl -L and a browser User-Agent and returned 200 unless noted.

---

✅ TO ADD (30)

Mode A — launched or majorly released in window (7)

1. DeepSeek Harness (deepseek-harness) — https://deepseek.com/harness
   Plugin-composed agent harness: models, tools, skills, sessions, sandboxes, storage and UI are all swappable plugins.
   Signal: 102,609★ verified — repo created 2026-08-13, i.e. the whole star count accrued in ~48 hours. Official deepseek-ai org. PH Weekly drop 2026-07-25: 28 added, 0 updated #6 on Aug 14 (156 upvotes). A plugin ecosystem already formed around it in-window (modlens 1,459★, iPolloWork 4,051★, OpenBiliClaw 2,436★ all citing DSH plugin support). Marked featured — the only one this run.

2. Muse Code (muse-code) — Meta Superintelligence Labs
   Terminal agent for long-horizon multi-file work, with persistent async background sub-agents.
   Signal: Meta launch 2026-08-05; PH Submission: Vecbase — Your AI team for company work. Product: https://ve… #4 on Aug 6 (248 upvotes); TechCrunch + The Register coverage. ⚠️ Caveat: the product host dev.meta.ai returned HTTP 500 to my UA. I used the research.meta.ai launch post (200) as agentUrl and left verified off. Your call whether that clears the bar.

3. Xirp (xirp) — Spotify, https://xirp.spotify.com/
   Manages concurrent Claude Code / Codex / Gemini CLI sessions, each in its own git worktree, context surviving harness switches.
   Signal: PH Submission: E2E TEST (from Claude, safe to close): verifying the /submit… #3 on Aug 11 (282 upvotes); HN front page Aug 10; Spotify Engineering states 1,300+ internal engineers on it pre-beta. verified left off — closed beta, marketing page only.

4. BrowserOS (browseros) — https://browseros.com
   Local open-source Chromium that external agents drive against your already-logged-in sessions via compact page snapshots.
   Signal: 13,202★ verified, pushed 2026-08-15. The "neo" release was the in-window hook (PH feat: add Umami traffic analytics #5 Aug 7–8, 209 upvotes) but this is a pre-existing project the catalog missed entirely. Name-collision checked: distinct from browser-use, browserbase, browseract, browser-harness.

5. Dograh (dograh) — https://www.dograh.com
   Self-hostable open-source voice-AI agent platform with visual workflow builder, telephony and BYOK pipelines.
   Signal: 5,339★ verified; PH 💀 Link-rot report 2026-07-20 — dead links in directory #1 of the day on Aug 12 with 542 upvotes. Distinct from catalog's vapi / retell-ai — this is the open-source competitor.

6. Kiro Crew (kiro-crew) — Kiro (AWS), https://kiro.dev/crew/
   Persistent workspace carrying dev context across sessions, coordinating agents over CLI, chat and remote SSH.
   Signal: 2,890★ verified in under a month on an AWS-published repo, pushed 2026-08-15. PH Weekly drop 2026-07-25: 28 added, 0 updated #6 on Aug 5 (175 upvotes).

7. Superlog (superlog) — Pulsent Labs, https://superlog.sh
   Reads Sentry/Datadog errors, reproduces them, opens PRs with fixes; memory trained on prior PR review feedback.
   Signal: 1,251★ verified; YC-backed; PH Submission: E2E TEST (from Claude, safe to close): verifying the /submit… #3 on Aug 6 (287 upvotes). Note: thematically adjacent to catalog's sentry-ai and retrace, but a separate vendor with a distinct product.

Mode B — established open-source gaps (11)

8. TradingAgents (trading-agents) — Tauric Research — 98,218★ verified, pushed 2026-07-18. Multi-agent LLM trading firm simulation (analyst/researcher/trader/risk roles). The reference agentic-finance repo and the single largest OSS gap in the catalog. Note: repo homepage field points at an arXiv PDF, so I used https://tauric.ai (200) as providerUrl and the repo as agentUrl.
9. MinerU (mineru) — OpenDataLab / Shanghai AI Lab — 77,663★ verified, pushed 2026-08-14. PDF/Office → LLM-ready Markdown. Catalog had docling / markitdown / unstructured but not the biggest one.
10. PageIndex (pageindex) — Vectify AI — 35,189★ verified, pushed 2026-08-14. Vectorless reasoning-based tree retrieval; the prominent anti-vector-DB RAG alternative.
11. ZeroClaw (zeroclaw) — ZeroClaw Labs — 32,584★ verified, pushed 2026-08-14. Small fully-autonomous assistant runtime. Name-collision checked against catalog's openclaw / clawvisor / clawk — different product.
12. Beads (beads) — Gastown Hall — 26,327★ verified, pushed 2026-08-15. Graph-based issue/memory store giving coding agents durable cross-session working memory.
13. A2A Protocol (a2a-protocol) — Linux Foundation — 25,350★ verified, pushed 2026-08-14. Structural gap: the catalog covers MCP heavily and had zero A2A entry.
14. MCP Python SDK (mcp-python-sdk) — Model Context Protocol — 24,005★ verified, pushed 2026-08-14. Catalog listed FastMCP (the derivative) but not the official SDK.
15. PentAGI (pentagi) — VXControl — 21,831★ verified, pushed 2026-08-06. Autonomous multi-agent penetration testing.
16. Eigent (eigent) — Eigent AI — 15,006★ verified, pushed 2026-08-14. Local multi-agent desktop workforce.
17. VoltAgent (voltagent) — 10,361★ verified, pushed 2026-08-10. TypeScript agent framework + observability console; clear gap next to Mastra and the Vercel AI SDK.
18. Agent S (agent-s) — Simular AI — 12,155★ verified, pushed 2026-08-01. Computer-use agent with hierarchical planning and GUI grounding; top OSWorld name. Catalog had cua and ui-tars-desktop but not this.

Mode B — established commercial platforms (8)

19. Mistral AI (mistral-ai) — the biggest lab-tier gap in the catalog. OpenAI, Anthropic, Google, Groq and Together were all listed; Mistral was not. Agents API with persistent agents, connectors and MCP client support. €1.7B Series C at €11.7B.
20. Sierra (sierra-ai) — Bret Taylor's conversational-agent company. $950M Series E at $15.8B (May 2026). Declarative Agent SDK is the developer surface.
21. Decagon (decagon) — $250M Series D at $4.5B (Jan 2026), 3x jump from $1.5B in Jun 2025.
22. Legora (legora) — $600M Series D at $5.6B (Mar 2026, Accel-led). Its aOS layer is an MCP client — customers build MCP servers against it. Catalog had harvey-ai and casetext-cocounsel but not Legora.
23. Replit Agent (replit-agent) — $400M Series D at $9B (Mar 2026). Catalog had cursor / cline / windsurf / devin but no Replit.
24. Factory (factory-ai) — $150M Series C at $1.5B (Apr 2026, Khosla-led). The droid CLI has MCP support and a headless droid exec for CI.
25. Warp (warp) — 64,224★ verified, pushed 2026-08-15. Rust terminal + Oz orchestration dispatching concurrent cloud agents. Note: the standalone Warp Agent CLI (launched Aug 4, HN 111 pts) is covered by this entry rather than split out — flag if you'd rather have two.
26. Baseten (baseten) — $1.5B Series F at $13B (Jun 2026). Catalog had Groq / Together / Fireworks / Modal / Replicate but not Baseten.

Mode B — official vendor MCP servers (4)

27. dbt MCP (dbt-mcp) — dbt-labs/dbt-mcp, 596★ verified, pushed 2026-08-14. Semantic Layer + Discovery API + dbt CLI. Catalog had Snowflake/Databricks/ClickHouse/Airbyte but no dbt.
28. Coinbase AgentKit (coinbase-agentkit) — coinbase/agentkit, 1,282★ verified, pushed 2026-08-13. Onchain wallets for agents; underpins x402.
29. Canva MCP (canva-mcp) — first-party hosted MCP at mcp.canva.com. Catalog had Figma twice and no Canva.
30. Tableau MCP (tableau-mcp) — tableau/tableau-mcp, 322★ verified, pushed 2026-08-15. The catalog had no BI-tool MCP at all.

---

🔄 UPDATED (0)

No field updates this run. I checked the one flagged candidate — stripe-agent-toolkit. Its repo moved (stripe/agent-toolkit → stripe/ai, 1,747★, pushed 2026-08-15), but the entry's agentUrl is https://docs.stripe.com/agents, which returns 200 and is still canonical. Nothing factually wrong, so I left it alone.

Non-blocking observation, not applied: that entry has only 3 tags (payments, api, agent-toolkit) against the 6–8 house standard. That's a quality gap rather than a factual error, so it stayed out of scope for this PR.

---

⏭ REJECTED (selected — full sweep was much larger)

Failed the programmatic-access requirement (web-UI only): Soloop (PH #2 Aug 7, 485 upvotes) · Rindler (PH #3 Aug 7) · Discovered Materials (YC P26, HN 159 pts) · Viktor.com · Hey Noah · Atlaso · Oasis · Human Behavior.

Models and research releases, not agent tooling: Muse Glimmer (HN #1, 1,203 pts — it's the model; Muse Code is its agent surface and is included) · Gemini 3.7 Flash · Maple-Preview · WorldClaw (Tencent) · Toast 1 embedding models · microsoft/fara.

Below the signal bar: mcp-stama (73 HN pts but 6★ verified and zero comments — vote pattern not credible) · mcptoon (144★, anonymous single-author) · fellowgeek/mcp-memory (152★, repo created same day) · Mole (126★) · Hax (223★, HN 112 pts — genuinely interesting C implementation, just under bar) · book-to-skill (1,080★ but 24 hours old, no homepage) · OAK/MasterAgent (335★, unverified benchmark claims) · Brandfetch/ZapDigits/BackEngine MCPs (mid-page PH, no cross-listing).

Verification failures:

• AgenticSeek (26,837★ — would otherwise have made the cut) — agenticseek.tech serves HTTP only; the HTTPS endpoint fails to connect. Individual maintainer, and catalog already has openmanus / suna / openhands in the local-Manus niche. Dropped on the TLS failure.
• Getty Images MCP (announced Aug 12, NYSE: GETY) — real first-party vendor MCP, but sales-gated with no public endpoint or docs. No self-serve programmatic surface to verify.
• Hebbia — $130M Series B at $1B, but docs.hebbia.ai fails DNS resolution. Could not verify a public API.
• ServiceNow AI Agents — product page returned HTTP/2 INTERNAL_ERROR across repeated attempts. Worth a manual retry next run.
• Zenity — $125M Series C (Aug 3, in-window, Norwest-led with SoftBank Vision Fund 2) is an unambiguous signal, but its docs describe security-team workflows, not a developer API. Could not confirm a self-serve programmatic surface.

Stale (fails the Mode B 12-month commit rule): AgentGPT · SuperAGI · BrowserMCP · vocode-core · LaVague · ragapp · codeinterpreter-api · stackblitz/bolt.new (last push 2024-12-17) · PostHog MCP (archived 2026-01-19) · box-community/mcp-server-box (archived, superseded by Box's hosted first-party server) · Algolia MCP (official org repo has 0★, last push 2026-01-14).

Already in catalog (name-collision checks done): Ante · Needle2 · Docker Sandboxes · OpenChamber · Kitesurf · Cloudflare OS · Channels SDK · Prime Agent · HyperProbe · Nightcrawler · Armature · Cherry Studio · CowAgent · qm · harvey-ai · glean · deer-flow · openclaw · wonderful-ai · Clay · 11x · awslabs/mcp (= aws-mcp) · mcp-use/mcp-use (= manufact).

---

💡 CATEGORY PROPOSALS

None. No proposed category had 3+ qualifying candidates behind it.

Closest call: agent-payments — Coinbase AgentKit, PayPal Agent Toolkit and Cloudflare Wallets all appeared this run, and the existing commerce-payments (11 agents) absorbed them fine. Worth revisiting if x402-style rails keep landing. Flagging, not proposing.

---

📋 DEFERRED QUEUE (verified, cut for count discipline — first candidates next run)

These all cleared the bar and were cut only to stay inside the 10–30 target, not on quality:

Cerebras (Nasdaq IPO May 2026, $95B day-one) · Parallel Web Systems ($100M Series B at $2B, Sequoia) · Cohere ($970M raised) · Lovable ($200M Series B at $2.8B, official MCP) · Cartesia ($191M raised, Sonic-3.5 #1 on Speech Arena) · Writer ($1.9B) · Augment Code ($227M Series B) · Amp / Sourcegraph · Hindsight (19,969★) · AgentMemory (27,025★) · Memvid (16,217★) · memU (14,309★) · Memori (15,980★) · MCP TypeScript SDK (13,170★) · Bisheng (11,860★) · CubeSandbox (11,106★, Tencent Cloud) · HexStrike AI (11,011★) · Vision Agents (8,084★, GetStream — the catalog still has no vision-agent entry) · Agent Squad (7,738★) · Swarms (7,049★) · Asana MCP · monday.com MCP · Square MCP · PayPal Agent Toolkit · Auth0 MCP · Dynatrace MCP (the notable observability hole) · Tines · ngrok AI Gateway · Coldtea (PH #1 Aug 7, 518 upvotes) · Ito (PH #2 Aug 13, 439) · Kane CLI (PH #1 Aug 13, 459) · Mixedbread Toast 1 (HN 194 pts) · Phone Harness (1,784★) · Bullet (YC S26) · Hoplite (YC S26, native MCP server) · Paritok (1,328★).

The memory-category entries above were deliberately capped — adding 5 more would have swamped an already well-covered category.

---

🐦 X POST DRAFTS — do not post before this PR merges

⚠️ Timing: the 431 count is not true until this PR is merged. Post after merge, not before.

Option A — announcement (271 chars)

agentswitchboard.dev: 401 → 431 agents.

Gaps I'm embarrassed about: no Mistral. No Replit Agent. No A2A protocol entry at all.

Biggest launch: DeepSeek Harness. 102,609 stars, repo created 2 days earlier.

Every star count pulled live from the GitHub API, none scraped.

Option B — honest-gap framing (255 chars)

Shipped 400+ AI agents in a public catalog without ever listing Mistral AI.

Or Replit Agent. Or a single A2A protocol entry.

Fixed all three today. 431 now.

Also: DeepSeek Harness, 102,609 stars on a repo created 48 hours earlier.

agentswitchboard.dev

Option C — thread (5 posts)

1/

Weekly pass on agentswitchboard.dev. 401 → 431 agents.

The uncomfortable part isn't what launched this week. It's what I'd been missing for months.

2/ — launches

Launches worth your time:

DeepSeek Harness — 102,609 stars on a repo created 2 days earlier. Everything is a swappable plugin.

Xirp (Spotify) — runs Claude Code, Codex and Gemini CLI side by side, each in its own git worktree.

Dograh — 5,339 stars, open-source Vapi rival.

(276 chars)

3/ — commercial

The money moved too:

Sierra, $950M Series E at $15.8B
Replit, $400M at $9B
Baseten, $1.5B Series F at $13B
Legora, $600M at $5.6B
Decagon, $250M at $4.5B

All five were missing from the catalog. All five are in now.

4/ — the gaps

And the ones that were just embarrassing:

No Mistral AI, while OpenAI, Anthropic, Google, Groq and Together were all listed.

No A2A protocol entry, in a catalog full of MCP.

No BI-tool MCP at all until Tableau today.

Curation is mostly noticing what isn't there.

5/ — method

Method note, since catalog counts are cheap to inflate:

Star counts come from the GitHub API at write time, never memory, never scraped.

Every URL fetched before listing. AgenticSeek has 26,837 stars and got cut because its site is HTTP-only.

agentswitchboard.dev

Recommendation

Option B. The honest-gap framing outperforms announcements, and "shipped 400+ agents without listing Mistral" is a real, checkable admission rather than a humblebrag — it earns the DeepSeek Harness number that follows instead of leading with it.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
agentswitchboard.dev	Ready	Preview	Aug 15, 2026 7:20am

[greptile-apps]

Greptile Summary

The PR expands the catalog from 401 to 431 agents and records the additions in the changelog.

• Adds 30 schema-conforming agent records across the existing catalog categories.
• Adds matching, newest-first changelog entries for every new agent.
• Marks DeepSeek Harness as the featured entry for this drop.

Confidence Score: 5/5

The PR appears safe to merge, with no concrete blocking or non-blocking defects identified in the added catalog data.

The new records satisfy the catalog’s schema and uniqueness constraints, and every added changelog entry resolves to the corresponding agent record.

Important Files Changed

Filename	Overview
content/agents/deepseek-harness.json	Adds the drop's featured agent with valid catalog metadata, categories, enums, and a unique identifier.
content/agents/a2a-protocol.json	Adds the catalog's A2A protocol entry using supported schema values and existing categories.
content/agents/mistral-ai.json	Adds Mistral AI with valid orchestration and language metadata and no identifier collision.
content/agents/tableau-mcp.json	Adds the Tableau MCP integration with valid data-analytics metadata and access-method values.
content/changelog.json	Prepends 30 matching changelog entries whose slugs and names resolve to the newly added agent records.

_{Reviews (1): Last reviewed commit: "Weekly drop 2026-08-15: added 30 agents" | Re-trigger Greptile}